From mboxrd@z Thu Jan 1 00:00:00 1970 From: Radim =?utf-8?B?S3LEjW3DocWZ?= Subject: Re: [PATCH] KVM: x86: call irq notifiers with directed EOI Date: Wed, 18 Mar 2015 21:16:06 +0100 Message-ID: <20150318201606.GC3309@potion.brq.redhat.com> References: <1426703902-16818-1-git-send-email-rkrcmar@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: linux-kernel@vger.kernel.org, kvm@vger.kernel.org, Paolo Bonzini , Marcelo Tosatti , chao.zhou@intel.com To: Bandan Das Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org 2015-03-18 15:37-0400, Bandan Das: > Radim Kr=C4=8Dm=C3=A1=C5=99 writes: > > kvm_ioapic_update_eoi() wasn't called if directed EOI was enabled. > > We need to do that for irq notifiers. (Like with edge interrupts.) >=20 > Wow! It's interesting that this path is only hit with Xen as guest. Linux doesn't use directed EOI ... KVM should fail with anything that depends on PIT, so probably only Xen bothered to implement it :) > I always thought of directed EOI as a "security feature" since broadc= ast > could lead to interrupt storms (or something like that) :) I think it is just an unpopular optimization for large systems. (With multiple IO-APICs: IRQ handler knows which ones need the EOI, but LAPIC doesn't, hence we avoid some useless poking if OS does it ... EOI interrupt storm happens because right after EOI, the IO-APIC can send another interrupt and real hardware is slow, so CPU manages some cycles before receiving the next one, but KVM works instantaneously.)